A Bayesian ordinal analysis of an AI-supported geometry-design item in a predominantly Intermediate Phase teacher-education sample

Abstract Broad awareness of artificial intelligence (AI) and familiarity with specific tools may provide different information about responses to a subject-specific survey item. This cross-sectional analysis examined a single-item ordered response to an AI-supported geometry-design statement, not a measure of design competence. The sample comprised 130 preservice mathematics teachers at one South African university, including 129 (99.2%) enrolled in Intermediate Phase Mathematics Education. The item asked about confidence but supplied five familiarity-style response anchors. Bayesian cumulative ordered-logit models estimated associations with a six-item AI-awareness composite and a five-item AI-tool-familiarity composite, accounting for year of study, prior training, and prior AI use. A one-standard-deviation increase in awareness was associated with higher response categories (median odds ratio [OR] = 1.82, 95% credible interval [1.23, 2.55]), as was tool familiarity (OR = 3.22, 95% credible interval [2.12, 4.62]). Adding familiarity improved expected out-of-sample predictive accuracy by 19.59 expected-log-predictive-density units (SE = 5.14). The interaction did not improve prediction. Interpretation is bounded by the shared response anchors and the homogeneous sample. The findings concern this item in this sampled setting and do not establish design competence or generalise across teacher-education programmes.